Sign in or 

Performance comparisons (older Xeon generations and AMD Opteron)
| Intel_GWagnon | Intel Xeon 5600 Series Processors: http://www.intel.com/itcenter/products/xeon/5600/index.htm |
| Intel_GWagnon | Wooohooo!!! |
| surbahns | Can't wait to hear about the Intel Xeon 5600 Series Processors! :) |
| Corey | Glad to be here. The links to the Xeon 5600 Series were helpful. |
| Dell-ScottH | Corey, you’ll have to thank Erson when he joins, he's great at digging up Internet gold. |
| Dell-ScottH | Hey Erson, we were just talking about you. Thanks for adding all those great links to the page. :-) |
| erson | No problem. |
| Dell-ScottH | I see we have some new faces, so I'll throw some helpful things out. |
| Dell-ScottH | First, have fun and don't worry about interrupting anyone. It's a chat, we can keep up. |
| Dell-ScottH | Next, there will be a transcript posted on this same link, so don't worry about writing down any of the links, they'll be here for you tomorrow. |
| Dell-ScottH | And if you see a link, be safe and right click on it or you might be kicked from the chat. It’s a weird client issue. |
| Dell-ScottH | If you do get kicked out, just use "action: recent room history" to catch up on what you missed. |
| Dell-ScottH | And with that I'd like to welcome everyone to this week’s chat, the launch of the Intel Xeon 5600 Series Processors, a.k.k the Westmere-EP family of processors. |
| Dell-ScottH | We have several talented people from Intel and Dell with us today. |
| Dell-ScottH | Please everyone introduce your self at one time! :-) |
| Dell-ScottH | This is when it starts to get fun; the intermingling of conversations is my favorite part. |
| Intel_GWagnon | Thanks for allowing us at Intel to join you today. We're happy to discuss our latest mutual offering around the Intel Xeon 5600 Series Processors. |
| Dell-JeffL | Hey! I'm Jeff Layton. I'm the enterprise technologist for HPC for Dell in the Americas. |
| Intel-ChadM | Hi. I'm Chad Martin, a field applications engineer for Intel based in Austin, Texas. |
| Dell-ScottH | Greg Wagnon runs the @intelxeon twitter account and the Intel Server Room community. |
| Corey | Corey McCormick with New Dominion, a small petroleum company in Oklahoma. |
| Dell-GarimaK | Hi, I'm Garima Kochhar, engineer with Dell's HPCC team. |
| jmcgehee | John Mcgehee with JDA Software Group, formerly i2 Technologies. |
| DELL-ScottP | Hi all. I am Scott Patti, one of the engineers who worked on the Westmere-EP effort here at Dell. |
| Intel-SteveT | Hi, I'm Steve Thorne, the product line manager for the Intel Xeon 5000 Series. |
| Dell-MonicaG | Hi I'm Monica Gonzalez, I am also one of the engineers who worked on Westmere-EP effort at Dell. |
| afuochi | Hello everyone! I'm Andre the Enterprise PR person at Dell. |
| Dell-ScottH | As we get going, feel free to kick out any questions. We'll weed through them and respond. |
| erson | Hi, I'm the guy who refreshed Intel’s homepage like crazy 12 hours ago to get all the links for this chat start page. :) |
| Corey | Thanks very much Erson! |
| Dell-ScottH | :-) |
| Dell-JeffL | LOL |
| surbahns | When can we expect the Intel Xeon 5680 to be in the R710 server line? |
| Dell-ScottH | First for the Intel folks, what are some of the great new features in this chip? |
| Intel-SteveT | We are very excited about the Xeon 5600 Series (aka "Westmere-EP"). |
| Dell-ScottH | I'd also like to learn more about the Trusted Execution Technology (TXT) features of these chips, where's the best place to learn more about that feature? |
| Intel-SteveT | It's based on Intel's 32 nanometer Logic technology. It allowed us to include 6-cores and 50% more on-board memory. |
| Dell-ScottH | Or maybe I'm getting ahead of myself and being greedy, but how many cores/threads? |
| Intel-SteveT | Up to six cores and 12 threads with Intel Hyper-Threading Technology. |
| Intel_GWagnon | And, all at the same or better power envelope. |
| Intel-SteveT | What resulted with the Xeon 5600 is either: up to 60% more performance or, if you held performance constant to Xeon 5500 Series (Nehalem-EP), a 30% power reduction. |
| erson | For more about performance go to: http://www.intel.com/performance/server/xeon/summary.htm |
| Dell-ScottH | From the Dell side, which new servers will feature this microprocessor? |
| Dell-ScottH | And is there anything important to note if I have an existing server and want to upgrade? |
| erson | The following will feature this new microprocessor: the entire line of two-socket PowerEdge servers, including two blade servers (M710, M610), four rack servers (R710, R610, R510, R410) and three tower servers (T710, T610, T410). |
| Intel_GWagnon | FYI: Dell's press release: http://content.dell.com/us/en/corp/d/press-releases/2010-3-16-poweredge-westmere.aspx |
| Corey | From those performance links it seems as though the biggest advantage was that you got to add most of the horsepower of an extra two cores without having to add the infrastructure for another server. The total wattage went up on the high end about proportional to the increase in the core count. |
| Dell-GarimaK | Existing servers need a Bios and iDRAC firmware upgrade to use the “Westmere-EP” processors. |
| Cesar | What is the status of obtaining Dell PowerEdge R610’s with the Nehalem-EP? |
| erson | Intel: How much did the power gating of uncore help to reduce power compared to going to 32 nanometers? |
| Dell-ScottH | Corey, I personally think this is a huge thing and will drive us away from the bigger and bigger multi-socketed servers that some of the other vendors are doing. |
| Corey | What about the workstation applications for the Xeon 5600? |
| Intel-ChadM | Corey, wattage actually stays the same if you look at the same frequency. The extra top end wattage was to get a higher frequency. |
| Corey | Scott, I agree. It is rare that I see the need for more than two sockets, and even 4 is on the edge of diminishing returns. |
| Dell-GarimaK | Corey, with some HPC applications too, total power consumption stayed about the same or was lower compared to Nehalem-EP. (12 cores vs. 8 cores) |
| Intel-SteveT | Erson: 32 nanometers provided a significant reduction in power consumption. It allowed us to maintain similar core frequency as Xeon 5500 Series (Nehalem-EP) at common TDPs. |
| erson | "While primarily aimed at workloads requiring systems of 4 sockets or more, there is an emerging interest in 2-socket systems based on the Intel Xeon processor 7000 class. Such systems will be in the market starting with the Intel Xeon processor 7500 series." |
| erson | That quote is from the server selection PDF from Intel. |
| DELL-ScottP | Cesar, Can you elaborate on your question regarding Dell PowerEdge R610’s and Nehalem-EP? |
| Intel-ChadM | Corey, We're excited about Westmere-EP in workstations. While some applications don't scale with cores, keeping frequency the same allows those applications to still perform well, enable more multi-tasking scenarios, and really scale up performance on high thread count workloads. |
| erson | Xeon 5600 also supports low power DDR3 which brings down the normal 1.5V to 1.35v. That will make a difference with all those memory sockets. |
| Dell-ScottH | Well, we'll just have to wait for the Xeon 7500 announcement then! :-) |
| jmcgehee | So, single threaded performance is the same as Nehalem-EP? |
| surbahns | When will we be able to configure systems on the Dell Web site with Xeon 5600 processors? |
| Dell-ScottH | Andre, Frank, when will they be in the online configurator for customers to order? |
| Corey | Chad, those are exactly the problems we face. Many of the workstation applications are clock limited because of the issues with developing efficient multi-threaded applications. |
| Intel-ChadM | Single-threaded performance is roughly the same as Nehalem-EP. It can be better when workloads can take advantage of the extra cache. |
| Intel-ChadM | Corey, Does more cache help your workloads? |
| Dell-FrankR | Cesar, Dell PowerEdge R610 offers Nehalem-EP today. New Nehalem-EP SKUs should be available next week with the Westmere-EP processors. Some Westmere-EP SKUs are available today. Please see Dell.com |
| erson | So, what is the Intel Turbo Boost Technology binning? I haven't found that information on the Intel homepage yet. |
| Corey | Chad, sometimes. I think the Turbo Boost Technology combines with the cache to help a bit. We are just now upgrading from the Intel Xeon WS CPU to try the next generation. Core i7 helps a bit. |
| Dell-FrankR | Surbahns, Westmere-EP processors will be generally available on Dell PowerEdge R710 next week. Two SKUs are available now. Please see Dell.com |
| surbahns | Great, thanks Frank. |
| Intel_GWagnon | Erson, the Turbo bins you get depend on the SKU. |
| erson | Yes, but what are they? |
| Intel-ChadM | Increasing cache will continue to be a knob we use in addition to some other tweaks in Westmere-EP. We also are working with workstation ISV's to help them get more threaded, but as you can imagine it's a long process. I think they are starting to see the need more now though. |
| Intel-SteveT | Erson, Turbo Boost improvements vary by processor speed and thermal design power (TDP). In general, one exception is an additional bin/133MHz for 130W TDP/6-core. The 95W TDP/6-core will give you 2 bins/266mhz. |
| erson | For the Xeon 5500 I could find that information in the Specification Update. |
| Corey | Chad, the Nehalem-EP is the first time in quite a while we have seen a significant increase in workstation performance per the money spent. We could always spend more money, but the returns were not proportional to the cost. |
| erson | So, for a 95W TDP/6-core it's perhaps 2/2/3/3/3/4? |
| Dell-FrankR | Selected Westmere-EP SKUs are in the online configurator now. More will be added early next week. |
| erson | It's was pretty much different for every model on the Xeon 5500 Series, so I assume it's the case with Xeon 5600 as well. |
| Corey | In looking at all the SKUs for these CPUs, it can be a pretty complex task to select the "correct" CPU for each application. |
| erson | Oh, I see that virtualization round-trip latency has been decreased by 12% compared to Xeon 5500 Series. |
| Intel-ChadM | Erson, 95W TDP/6-core is 2/2/2/2/3/3. |
| Dell-ScottH | It definitely has gotten more difficult from the days of "just get the biggest one." |
| erson | Chad, great, hope you release a document with that information. |
| Corey | I think all the virtualization enhancements are really improving our VM performance and scalability. |
| Intel_GWagnon | Not sure it is more difficult, as much as it is actually easier to find a solution that matches your needs. |
| Intel-ChadM | 60W T DP/6-core is 2/2/3/3/4/4. |
| Dell-JeffL | Having so many options means you have to understand your application and environment better (not always a bad thing). |
| erson | Corey, glad you got rid of the Westmere-EP models with dual CPU support though. |
| Dell-ScottH | So many different application profiles. Any plans from Intel on releasing a paper about best processor per type of application profile? |
| Corey | Greg, not when you factor in costs/pricing that is not linear. |
| Intel_GWagnon | Per workload, it will take some research since many people run multiple workloads on a given server. |
| erson | It looks like Westmere-EP models this time will only be Up, correct? |
| Er1cb | There'll be more of an emphasis to write code that takes advantage of newer processors. |
| Dell-JeffL | Greg, definitely true. |
| Intel_GWagnon | Then there are those that only run one workload per server, and so it continues. |
| Dell-ScottH | Guess that's why we all still have jobs in technology! |
| Intel_GWagnon | Choosing a price point, and general power efficiency level that you might want is always a good place to start. |
| Intel-ChadM | Erson, that's correct because the Dell 130W DP models are now supported on servers in addition to workstations. |
| Corey | Absolutely, kind of like surfing. |
| Dell-JeffL | In the HPC world we have pretty good tools for characterizing applications, but it takes work and if you have a bunch of them life is very interesting. :) |
| Dell-GarimaK | The processor finder on Intel’s Web site lists Turbo Boost bins for Nehalem. Maybe Westmere-EP details will get added there in time. http://processorfinder.intel.com/list.aspx?procfam=528&sspec=&ordcode= |
| erson | Chad, those are great Turbo Boost binnings. Definitely an improvement to the first generation of binnings on the first Intel Core i7s. |
| Intel_GWagnon | Lv SKU's (l5600) are rather good at performance and low power as well. |
| erson | Garimak, they should put the information on ark.intel.com as well in that case. |
| Corey | I am looking forward to keeping our server room HVAC alone for a while. Adding CPU without adding A/C will be great! |
| erson | I'm surprised you haven't mentioned Intel Advanced Encryption Standard - New Instructions (AES-NI) yet. Tell us more about it. |
| Dell-ScottH | Corey, I think many people are very happy with that right now. |
| Intel_GWagnon | Erson, shhh! AES-NI is encrypted and therefore secret. ;o) |
| Intel-SteveT | Erson, we expect good things from AES-NI (Intel Advanced Encryption Standard - New Instructions). These new instructions allow many more uses for AES, whether it's in full disk encryption or SSL transactions. |
| surbahns | Thank you for the update. I can't wait to order some Dell PowerEdge R710s next week with the Intel Xeon 5680s. See ya! |
| Intel-ChadM | The exciting thing about AES-NI for me is HTTPS with Westmere-EP >20% faster that vanilla HTTP with Nehalem-EP. |
| Corey | The iSCI checksum instructions in the Xeon 5500 CPU really changed our iSCI world. Bad for the iSCI HBA vendors, but good for our wallets. It seems that would be a big win if it trickle down to the Cpus used in iSCI SAN targets. |
| Intel-SteveT | Erson, we built some technology demonstrations based AES-NI that show significant improvement in performance of encryption/decryption when taking advantage of those new instructions. Quite a few software vendors are supporting them, including MacAfee, Microsoft, Oracle and VMware. |
| Corey | Are the new AES-NI instructions supported today on existing code? |
| Intel-SteveT | Surbahns, thanks for your participation and positive comments! |
| Intel_GWagnon | An article from InfoWorld.com this morning showed 400% improvement on his test with encrypting a large file using AES-NI enabled/disabled. |
| Intel-SteveT | Corey, the existing AES-code needs to be modified to take advantage of the new instructions. Intel provides a number of compliers and tools to accomplish this. |
| erson | BitLocker supports AES_NI. |
| Intel-ChadM | AES-NI was communicated to ISVs more than a year ago. OpenSSL supports it, although and I don't know which version, and GCC 4.4.0 supports it as well. |
| Dell-ScottH | Here’s the article: http://www.infoworld.com/node/116588 |
| Corey | Stevet, thanks. I wasn't sure how long it would be before we see the benefits. |
| Corey | Erson, I didn't know that. That is really good news. |
| surbahns | Quick question if I may. If you compare the Xeon 5570 and Xeon 5680 (http://ark.intel.com/compare.aspx?ids=37111,47916) you will see the Xeon 5680 doesn't have virtualization technology support. Why is that? |
| Intel-ChadM | Corey, the CPU utilization improvement with BitLocker is pretty dramatic. |
| erson | Also, Xeon 5600 supports for 18x16 GB while Xeon 5500 "only" supports 12x16 GB. |
| Intel_GWagnon | 'only'... :) |
| erson | Maybe I should have typed """only"""" |
| Corey | I think the lower power memory should really help keep things cooler. It gets pretty hot here in the summer. I don't think you can ever have too much RAM. |
| Intel-SteveT | Surbahns: I think that is a typo in ark.intel.com. Xeon 5680 certainly supports virtualization technology. |
| erson | "According to Samsung, 48 GB of 40nm low power Ddr3 1066 should use on average about 28W (an average of 16 hour idle and 8 hours of load). This compares favorably with the 66W for the early 60nm Ddr3 and the currently popular 50nm based Dram which should consume about 50W. So in a typical server configuration, you could save – roughly estimated – 22W or about 10% of the total server power consumption." |
| Intel_GWagnon | I know the ark.intel.com guys. I will make sure they get it cleaned up. Thanks for pointing it out. |
| Corey | Erson, we are at our thermal limits for the most part and in order to add either more SAN or more CPU, something has to give. |
| Corey | The Xeon 5600 Series seems like a good step in the right direction. |
| Intel-SteveT | Greg, please have them fix for all Westmere-EP SKUs. Virtualization technology is supported on all of them, but for some reason ark.intel.com doesn't reflect properly. |
| erson | Intel Xeon also supports two Dimms at 1333mhz, while the Xeon 5500 would throttle down to 1066mhz. Although I have seen that bios a couple of months ago made 1333mhz possible for two Dimms with the Xeon 5500 as well. (perhaps not supported by Intel though?). |
| surbahns | With 10GB iSCSI NICs we prefer to use VMDirectPath that uses the Vt-d. So I'm glad to hear this is a typo. |
| Corey | We can consolidate several of our older Dell PowerEdge 2850s into just a couple of Dell PowerEdge R series and save 500-2000 BTU each. |
| Intel-SteveT | Erson, 2 DIMMs per channel at 1333mhz is supported by Intel, but for Intel Xeon 5600 only. |
| Corey | Surbahns, which 10 iSCSI NICs are you using? |
| Intel_GWagnon | I already sent the request. |
| erson | Yeah, seems like HP and Dell support 2 DIMMs per channel at 1333mhz with Xeon 5500 when using the latest bios though. |
| Dell-JeffL | Erson, that's correct. |
| surbahns | Corey, they are Intel, let me get the model number. |
| erson | Stevet: three DIMMs per channel still drops it down to 800mhz or just 1066mhz? |
| Corey | We have the Intel 10G, but they do not offload iSCSI AFAIK. |
| Intel-SteveT | Erson, unfortunately max speed at three DIMMs per channel is only 800mhz. |
| Corey | Stevet, so is the 18 x 16G 3 channels or ??? |
| erson | Corey: yes |
| erson | 1 channel = 6 DIMMs, 2 channel = 12 DIMMs, 3 channel = 18 DIMMs |
| Corey | Erson - so 6 DIMMs is the max for memory bandwidth? |
| erson | Hmm, poor wording by me there. |
| erson | Each CPU has three memory channels and each channels supports up to three DIMMs. |
| Dell-GarimaK | Corey and Erson, Memory bandwidth is good with 2 DIMMs per channel too. So 12 DIMMs total. |
| erson | Corey, Yes, if your workload needs memory bandwidth you should only populate up to 12 DIMMs. |
| Corey | I see. 2 DIMMs x 3 Channels x 2 CPU |
| erson | Since then you can use 1333mhz DIMMs (if your CPU supports it). |
| Corey | It would be awesome if all this is spelled out in the technical notes for the Dell servers... :-) |
| Dell-JeffL | Corey, just be sure to get 1333 Mhz DIMMs and make them the same size if you want to squeeze every last ounce of memory bandwidth (also use RDIMMs). |
| Dell-GarimaK | Corey, that's right, for 1333mhz |
| erson | You need the X-models to be able to utilize 1333mhz DIMMs, right? |
| Corey | Excellent info. |
| Dell-ScottH | Corey, Jeffl had a great blog post on it for the first Xeons, Jeff you have anything planned, or can point back to that blog? |
| Intel-SteveT | Erson, yes, typically the Intel Xeon 5600 processors with an "X" product number support the fastest memory speeds. |
| Dell-JeffL | Nothing new planned, so the old blog is still good. |
| erson | Anything with a 6.4GT/s QPI supports 1333mhz DIMMs |
| Dell-GarimaK | Corey, our team did a white paper on Memory guidelines for HPCC applications for Nehalem-EP. This was with the first Bios we had, so data is a little old now. But you might be interested in it. http://content.dell.com/us/en/enterprise/d/business~solutions~whitepapers~en/documents~11g-memory-hpc-wp.pdf.aspx |
| Intel-ChadM | The 1333mhz DIMMs exception is the Intel Xeon L5640. Other than that it's the Xeon and Westmere-EP SKUs. |
| Intel_GWagnon | Corey, regarding Intel 10GB support for iSCI, ping them about it at http://twitter.com/intelethernet. I believe they do support it. |
| Dell-JeffL | http://www.delltechcenter.com/page/04-08-2009+-+nehalem+and+memory+configurations |
| Intel-SteveT | Erson, not always the case. Intel Xeon L5640 also supports 1333mhz memory, and it only has 5.86GT/s Qpi |
| Corey | Not all the Xeon though seem to do 1333mhz. |
| Dell-ScottH | Jeff you beat me to the copy paste! :-) |
| Corey | Greg, thanks. |
| Intel_GWagnon | No problem. |
| Dell-ScottH | We're coming up on the end of the hour and I know some will have to drop so I wanted to thank everyone before that happened. |
| Dell-ScottH | Of course, anyone/everyone is welcome to stay as long as they wish. |
| Intel-ChadM | Corey, on my chart all of the Xeon SKUs list 1333mhz. Which one are you seeing show up differently? |
| erson | Glad it's a session on Thursday as well. It feels like we just started. |
| Dell-ScottH | Thanks to all the Intel and Dell teams for joining. |
| Dell-ScottH | Yes, please join us this Thursday at same time for continuation of this chat |
| Intel-SteveT | Thanks for inviting us today. Please send all of your most difficult questions to Greg Wagnon. ;-) |
| Dell-ScottH | :-) |
| erson | Deep Diiiive |
| Dell-ScottH | We’re diving deeper on Thursday! |
| Intel_GWagnon | Appreciate the questions and the time. Feel free to send questions to the @Intelxeon account on Twitter and I will track down some answers. |
| Corey | Chadm, http://www.intel.com/assets/en_us/pdf/prodbrief/323501.pdf |
| Corey | Jeffl, thanks. This link though says that the 1333mhz is for only one DIMM. |
| Dell-JeffL | Corey, which link? |
| Dell-GarimaK | Corey, Keffl...the original memory article you did, Jeff. |
| Dell-JeffL | Corey – oops, you're correct. That was before the updated BIOS was released. I need to edit that blog, don't I? |
| Dell-GarimaK | That was with the initial BIOS. |
| Corey | Ok, thanks. |
| Dell-GarimaK | Starting BIOS 1.1.4 for Nehalem-EP, Dell does support 2DPC @ 133mhz? |
| erson | Support for 1GB pages... hoho. |
| Dell-ScottH | Thanks to everyone ... I have to take off, but feel free to talk amongst yourselves and see you again in two days. |
| Intel-ChadM | Corey: I think that PDF is mistaken. I'll go ask that the team to look at it. |
| erson | Could be handy in when my server has a couple of TB of RAM. |
| Corey | Scott - Is there another on Thursday? |
| erson | Corey: same time, same bat-channel. |
| Corey | Thanks to you all. This was very informative. |
| erson | http://www.delltechcenter.com/page/03-18-10+intel+westmere-ep+deep+dive+chat |
| Dell-ScottH | Thank you Corey, appreciate the questions and participation. |
| erson | "One six-core Xeon 2.93 is worth two six-core Opterons at 2.6 Ghz." |
| erson | That's the VMmark result on AnandTech. |
| Corey | Erson, that link has a broken calendar item. |
| Corey | It is a recurring link for 2009 only. |
| erson | You will find the link on the front page tomorrow. |
| erson | Or you can just find it in the left side menu under "Blog Chats and RSS" and "Chat Topics and Transcripts" |
| Corey | Thanks, I was just letting someone know. It has the right description, but the .ICS file is not correct that is generates... same problem with the one today. |
| erson | On the 30th we have a chat about Nehalem-EX. |
| erson | I don't use the calendar ICS-files. I just follow @Dellservergeek on Twitter. :) |
| erson | I love the idle power of Xeon 5600. |
| Corey | Indeed... thanks again. Our servers really spend 12+ hours that way. The 5600 should really help. |
| erson | Hmm, Intel NICs doesn't have iSCI offload to the best of my knowledge |
| Corey | I thought the same. |
| Intel_GWagnon | Take a look at Intel Xeon L5640. I have that running in a white box system and total system idle is ~58w. Very nice. |
| erson | It does, however, support iSCI boot. |
| Corey | The 10G NIC works great, but no iSCSI. On most machines we don't care as the W2K8 iSCSI client uses the checksum and so performance is not a big deal. On VMware though we need the CPU back. |
| Dell-d_glynn | Your CPU-bound rather then memory-bound in your VMware enviroment? |
| erson | Ahh, the product brief for the latest 10GbE adapters with 82599 says that the latest Intel Xeon support the CRC instruction set which makes CRC calculation a lot easier. |
| erson | http://www.intel.com/assets/pdf/prodbrief/322217.pdf |
| Corey | D_glynn, not during the daily workloads.Tthe users are easy. |
| Corey | D_glynn, it is during our maintenance, backups, antivirus scans, defragmentations, etc. |
| erson | I wonder if the 32NM quad cores are Hexacores with two cores disabled or if it's an entirely own design. |
| Corey | Erson, a great way to salvage a chip with a bad CPU... |
| erson | Yeah, common practice to increase yields. |
| Corey | Erson, I would guess you are correct. Each CPU is worth a great deal in revenue if it can be salvaged. Yields are really tough at the start of a newer process I am told. |
| Dell-d_glynn | Corey, then when is it that you are lacking in CPU, and feel that NICs with iSCSI offload would help? |
| erson | and by the way, there is an Intel Xeon W3680 released today as well for single socket workstations. |
| erson | It has the same price and specifications as the consumer Core i7 980X, $999. |
| Corey | That could help a lot. Much better price than some of the other workstation CPUs. |
| erson | I better not tell my developers about that one or they will be begging hard! |
| Corey | D-glynn, during our CPU/I/o intensive windows. Of course we would all be much better off with 3 x 6ghz, than 6 x 3ghz, but thermally that doesn't work today. Mostly it is our dependence on single/limited threaded applications. The schedulers do move things around, |
| erson | Power consumption of an entirely idle Xeon 5680 is ~6.3. |
| Corey | But the more predictable response of the HBA makes for smoother I/o. |
| Dell-d_glynn | I see, thanks Corey. |
| Corey | D-glynn, filling up the I/o for an LTO-4 is pretty hard. It sustains ~120MB/second before hardware compression. That is a heck of a stream to maintain. |
| erson | The fastest CPU for single threaded applications would be the Intel Core i5-670. |
| Corey | D-glynn, One example would be to compare the I/o numbers from FC-HBA with iSCSI. The aggregate numbers can look great, but it takes a lot of work to get the pipe flowing. |
| erson | It Turbo Boosts up to 3.73ghz and has 2MB per core (it's a dual core with HT). |
| erson | It's also a Westmere-EP. |
| Corey | Erson, you might think so but check the SPECint number and it does not work that way. |
| erson | Why is it slower? |
| Corey | It is. Significantly. |
| Corey | Sorry, are you asking me why? |
| erson | Yes |
| Corey | For a single desktop the i870 was king last I looked. I don’t know about the 980, but the 9-series was more for a CPU upgrade than the entire machine costs. |
| Corey | I think it had to do with the memory/cache architecture when I looked into it. They also changed something in the branch predictor maybe... that brain cell is not high confidence though. |
| erson | Yes, but weren't we talking about single thread performance which would translate to throttling down all cores but one? |
| Corey | Yes, but SPECint is for a single CPU, but SPECint Rate was the aggregate. |
| erson | Do you mean a single core? |
| Dell-d_glynn | Corey, it sounds like you are expecting an iSCCI offload card to perform akin to an FC-HBA, am I reading you right? |
| erson | Dell has a good partnership with Emulex if you want a good hardware iSCSI offload NIC. |
| erson | The broadcoms do it as well. |
| Dell-d_glynn | Is it supported by VMware? |
| erson | Emulex? |
| Dell-d_glynn | Yup |
| Dell-d_glynn | I suppose I should know the answer to that. :) |
| erson | I don't know but it must be supported. Both HP and IBM got their own versions of the same NIC. |
| erson | It also has full hardware support for FCOE. |
| Corey | D_glynn, not exactly like it. We are trying to consolidate the number of physical connections to our servers. |
| erson | Corey, Xsigo is the thing for you then. |
| Corey | Erson, thanks for your input. |
| Corey | Erson, I will take a look. :-) |
| Corey | D_glynn, Two dual-port 10G NICs with iSCSI gives us bandwidth, and redundancy that is enough for pretty much any application, VM, server, etc... |
| Corey | The new Dell 24-port 10G switches are a big enabler for that. Cisco 65xx switches are $10K per port (list). We can't do 10G x 4 per server. With the new 8024 and a reasonable 10G iSCSI NIC, we only need two x8 slots and all our I/o is covered. That means that 1U, or even 2/3U servers or blades can take care of everything. |
| Corey | Erson, FCOE is great, but doesn’t help the $/Gbps for storage. if you put FCOE in the description, the price goes up $1K per device/port. |
| Dell-d_glynn | There is always something newer on the horizon. |
| Dell-d_glynn | You seem to have a good grasp of your environment and it needs. |
| Dell-d_glynn | Right, time for me to head off. |
| Dell-d_glynn | Good luck, Corey. |
| Corey | Thanks all.... |
cbemily |
Latest page update: made by cbemily
, Mar 17 2010, 4:45 PM EDT
(about this update
About This Update
added transcript
- cbemily
4572 words added 24 words deleted view changes - complete history) |
|
Keyword tags:
Intel Westmere
More Info: links to this page
|